Look at mapping the interpreter's code in huge pages

[maximecb]

It looks as though the CPU has two pieces of iTLB hardware. One for huge pages and one for small pages. If that's true, we could benefit from having the interpreter use the huge pages, and the JIT use the small pages all for itself. This is a patch that we could also potentially upstream. It could benefit CRuby and MJIT as well.

See Intel doc and Intel sample implementation.

[maximecb]

@XrXr Since remapping the text segment is complicated. I think it make sense to focus on maybe just trying to mmap the microjit code into huge pages and test the performance for now.

We could leave text in normal pages, and have the JIT code in the huge pages, ideally allowing us to use both TLBs at once, an avoiding a complex, less portable solution.

If you try this, I would suggest using the code from the microjit-v1 branch, since the current branch is in flux and not ready to be benchmarked right now.

[maximecb]

Closing this for now since it doesn't look like mapping the text segment in huge pages is worth the added complexity.

Might be worth revisiting mapping the JITted code in huge pages (simpler) once MicroJIT V2 is further along.